Embedded Learning Based on RNN network and RotatE strategy for Relation Prediction in Knowledge Graph

摘要

Using knowledge graph-based relationship prediction methods usually uses a path-oriented strategy and ignores the differences between nodes neighborhood relations, resulting in excessive path redundancy information, which reduces the prediction accuracy. This paper proposes a method based on RNN network and RotatE strategy, which improves the accuracy of relationship prediction by embedding learning and accurately expressing the neighborhood relations of nodes. First of all, through the RNN network[9], the data association characteristics of different nodes' neighborhood are effectively learned, so that the parameters contain the information of the neighbor nodes; Then, the RotatE[5] score strategy is used to describe the differences in the relationship between the nodes, which makes the node relationships at the encoding end more clearly distinguishable. The strategy, which combines relationship and path information, uses the neighborhood similarity of nodes and path information as the main basis for relationship prediction, thereby improving the accuracy of relationship prediction. Experiments show that the method proposed in this paper is effective, and compared with the latest algorithm, its prediction accuracy has been significantly improved.